AI Token Spending Index drops nearly 20%, users' marginal willingness to pay may show signs of cooling

The global AI trading market is facing a cooling of a key observation indicator. The core index measuring AI token spending, after a rapid surge, has pulled back, prompting a reexamination of whether the AI cycle driven by massive capital expenditures remains stable—both demand intensity and pricing power are being subjected to stricter scrutiny.

According to Bloomberg, the LLM Token Expenditure Index compiled by Silicon Data nearly doubled since its launch in December last year, but has fallen nearly 20% after peaking in May. This index is regarded as the most direct market indicator of willingness to pay for AI services and changes in marginal demand, and its weakening is triggering a reassessment of the pace of AI commercialization.

Some investors believe this change may indicate that AI companies are entering a more cost-sensitive competitive environment, with pricing power potentially being eroded. Veteran investor Louis Navellier noted that there are already signs that some users are limiting their frequency of use due to cost pressures. Meanwhile, market rumors around OpenAI potentially delaying its IPO are also seen as indirect evidence of ongoing profitability challenges.

However, the decline in this index does not mean that AI services themselves are becoming "cheaper." Silicon Data emphasizes that this indicator is a composite result of the interaction between price and usage volume, acting more as a proxy variable for "marginal willingness to pay." Therefore, the driving factors behind its trend vary, and the corresponding market implications can be quite different.

Divergence Intensifies: Demand Cooling or Price Structure Shift

There is clear disagreement in the market regarding the explanation for the index decline.

One more optimistic interpretation holds that token prices have fallen by over 90% since 2023, significantly lowering the barrier to usage, thereby driving overall expenditure expansion. Under this framework, the periodic decline in the index reflects more of a restructuring of demand rather than an overall weakening of demand, and the logic of AI expansion remains intact.

However, a pessimistic view argues that this may indicate users' willingness to pay has approached a temporary ceiling. Allianz Research points out that the current growth gap between AI investment and actual sales is about 46%, higher than the approximately 32% deviation during the telecom bubble in 2001. In this context, any weakening on the demand side could amplify valuation pressures.

Louis Navellier also mentioned that companies are facing cost constraints in using AI services and are beginning to limit "unlimited usage," which some market participants see as an early signal of declining demand elasticity.

Capital Expenditure Logic Remains, but Structure Is Changing

Although demand signals are fluctuating, the AI infrastructure investment cycle has not clearly reversed. Market data shows that high-end GPUs and high-bandwidth memory are still in short supply, expected to continue until 2026, and possibly extending to 2028.

However, the market's focus is shifting from the training side to the inference side. The report indicates that this change means the structure of computing power demand is undergoing a transition: the share of high-end training GPUs may decline, while the demand for inference-optimized hardware rises relatively, thereby altering the beneficiary structure of the industry chain.

This change does not directly constitute a bearish judgment on the chip industry, but it may imply that the growth driver is shifting from "single high-end computing power expansion" to "structural demand redistribution."

Policy and Regulatory Variables Increase Pricing Complexity for Enterprises

In addition to demand itself, the regulatory environment is also affecting the commercialization path of AI products.

Recently, US regulators have requested adjustments to the release pace of certain models and relaxed access restrictions on some models from Anthropic PBC. Meanwhile, the EU AI Act imposes mandatory assessments and stricter transparency requirements on frontier models.

These policy changes do not directly limit prices, but they increase deployment and compliance cost pressures, making companies more inclined toward cost optimization when allocating workloads across different models. This trend may indirectly affect the pricing power and usage share of high-end models.

Risk Warning and Disclaimer

        Market risk exists, and investment requires caution. This article does not constitute personal investment advice, nor does it consider the specific investment objectives, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this content is at your own risk.